TWI660317B - Methods for predicting marketing target popularity and non-transitory computer-readable medium - Google Patents
Methods for predicting marketing target popularity and non-transitory computer-readable medium Download PDFInfo
- Publication number
- TWI660317B TWI660317B TW106145044A TW106145044A TWI660317B TW I660317 B TWI660317 B TW I660317B TW 106145044 A TW106145044 A TW 106145044A TW 106145044 A TW106145044 A TW 106145044A TW I660317 B TWI660317 B TW I660317B
- Authority
- TW
- Taiwan
- Prior art keywords
- keywords
- mentioned
- series data
- predicted
- marketing
- Prior art date
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一種行銷標的熱門度預測方法,包括:取得對應於一行銷類型之複數文章;取得對應於行銷類型之複數關鍵字;自上述文章取得對應於關鍵字之每一者之時序資料;根據每兩個關鍵字之時序資料計算對應的一相關性;根據相關性取得對應於每兩個關鍵字之一領先關係;根據領先關係以及關鍵字之每一者之時序資料建立一預測模型;根據預測模型以及一預測詞彙取得一推薦清單;以及根據推薦清單決定對應於預測詞彙之一行銷標的。 A method for predicting the popularity of a marketing target includes: obtaining a plurality of articles corresponding to a type of marketing; obtaining a plurality of keywords corresponding to a type of marketing; obtaining time series data corresponding to each of the keywords from the above article; according to each two Calculate a correlation for the time series of keywords; obtain a leading relationship corresponding to every two keywords according to the correlation; build a prediction model based on the leadership relationship and time series data for each of the keywords; based on the prediction model and A predicted list is obtained from a predicted word; and a marketing target corresponding to one of the predicted words is determined according to the recommended list.
Description
本發明係有關於一種行銷標的熱門度預測方法以及非暫態電腦可讀取媒體。 The invention relates to a method for predicting the popularity of a marketing target and a non-transitory computer-readable medium.
隨著社群媒體的高度普及化,越來越多企業透過社群媒體執行各種行銷策略。現有的利用社群媒體宣傳廣告的作法為偵測當前熱門的議題、對產品的相關詞語進行相關性分析、尋找與產品連動的相關議題,以作為下一波行銷標的之方向,例如購買和當前最熱門話題相關的關鍵字廣告或者邀請網紅部落客撰寫行銷文等。然而,現有的購買關鍵字行銷方式主要係著重於尋找熱門關鍵字的相關關鍵字,然而隨著使用熱門關鍵字的曝光度越高,該關鍵字的廣告成本也隨著增加。而有限的行銷預算下,若將所有預算僅投資於購買熱門關鍵字,其所帶來之行銷效益可能有限。因此如何找尋更佳的行銷標的以控制行銷預算為目前必須考慮之方向。 With the popularity of social media, more and more companies are implementing various marketing strategies through social media. The current practice of using social media to promote advertisements is to detect current hot topics, analyze the related words of products, and find related topics that are linked to the product as the next wave of marketing direction, such as purchase and current Keyword ads related to the hottest topics or inviting bloggers to write marketing articles. However, the existing marketing methods for buying keywords are mainly focused on finding relevant keywords of the popular keywords. However, as the popularity of using the popular keywords increases, the advertising cost of the keywords also increases. And with a limited marketing budget, if you invest all your budget only on buying popular keywords, it may have limited marketing benefits. Therefore, how to find a better marketing target and control marketing budget is the direction that must be considered at present.
本發明一實施例提供一種行銷標的熱門度預測方法,包括:取得對應於一行銷類型之複數文章;取得對應於行 銷類型之複數關鍵字;自上述文章取得對應於關鍵字之每一者之時序資料;根據每兩個關鍵字之時序資料計算對應的一相關性;根據相關性取得對應於每兩個關鍵字之一領先關係;根據領先關係以及關鍵字之每一者之時序資料建立一預測模型;根據預測模型以及一預測詞彙取得一推薦清單;以及根據推薦清單取得對應於預測詞彙之一行銷標的。 An embodiment of the present invention provides a method for predicting popularity of a marketing target, including: obtaining a plurality of articles corresponding to a type of marketing; obtaining a corresponding article Plural keywords of the type of pin; Get time series data corresponding to each of the keywords from the above article; Calculate a corresponding correlation based on the time series data of each two keywords; Get corresponding to each two keywords based on the correlation One of the leading relationships; establishing a prediction model based on the leading relationship and the timing data of each of the keywords; obtaining a recommended list based on the prediction model and a predicted vocabulary; and obtaining a marketing target corresponding to one of the predicted words based on the recommended list.
本發明另一實施例更提供一種非暫態電腦可讀取媒體,具有指令儲存於其中,當指令透過一電子裝置之一處理器執行時,致使電子裝置所執行之操作包括:取得對應於一行銷類型之複數文章;取得對應於行銷類型之複數關鍵字;自文章取得對應於關鍵字之每一者之時序資料;根據每兩個關鍵字之時序資料計算對應的一相關性;根據相關性取得對應於每兩個關鍵字之一領先關係;根據領先關係以及關鍵字之每一者之時序資料建立一預測模型;根據預測模型以及一預測詞彙取得一推薦清單;以及根據推薦清單決定對應於上述預測詞彙之一行銷標的。 Another embodiment of the present invention further provides a non-transitory computer-readable medium with instructions stored therein. When the instructions are executed by a processor of an electronic device, operations performed by the electronic device include: Plural articles of the marketing type; obtain plural keywords corresponding to the marketing type; obtain time series data corresponding to each of the keywords from the article; calculate a corresponding correlation based on the time series data of each two keywords; according to the correlation Obtain a leading relationship corresponding to every two keywords; establish a prediction model according to the leading relationship and the time series data of each of the keywords; obtain a recommendation list according to the prediction model and a predicted vocabulary; and determine a correspondence corresponding to the recommendation list according to One of the predicted words mentioned above is marketed.
100‧‧‧系統架構 100‧‧‧System Architecture
110‧‧‧處理單元 110‧‧‧processing unit
120‧‧‧儲存單元 120‧‧‧Storage unit
130‧‧‧網路介面 130‧‧‧Interface
140‧‧‧顯示單元 140‧‧‧display unit
150‧‧‧輸入單元 150‧‧‧ input unit
200‧‧‧行銷標的熱門度預測方法 200‧‧‧ Popularity Prediction Method of Marketing Target
S201~S209‧‧‧步驟流程 S201 ~ S209‧‧‧step flow
第1圖係顯示根據本發明一實施例所述之行銷標的熱門度預測系統之方塊圖。 FIG. 1 is a block diagram showing a marketing popularity prediction system according to an embodiment of the present invention.
第2圖係顯示根據本發明一實施例所述之行銷標的熱門度預測方法之流程圖。 FIG. 2 is a flowchart illustrating a method for predicting popularity of a marketing target according to an embodiment of the present invention.
第3圖係顯示根據本發明一些實施例所述之對應於不同關鍵字之時序資料之示意圖。 FIG. 3 is a schematic diagram showing timing data corresponding to different keywords according to some embodiments of the present invention.
第4圖係顯示根據本發明一實施例所述之兩個關鍵字之間之領先關係之示意圖。 FIG. 4 is a schematic diagram showing a leading relationship between two keywords according to an embodiment of the present invention.
第5圖係顯示根據本發明一實施例所述之時序關聯矩陣之示意圖。 FIG. 5 is a schematic diagram showing a timing correlation matrix according to an embodiment of the present invention.
有關本發明之行銷標的熱門度預測方法以及非暫態電腦可讀取媒體適用之其他範圍將於接下來所提供之詳述中清楚易見。必須了解的是下列之詳述以及具體之實施例,當提出有關行銷標的熱門度預測方法以及非暫態電腦可讀取媒體之示範實施例時,僅作為描述之目的以及並非用以限制本發明之範圍。 The method for predicting the popularity of the marketing target of the present invention and other areas applicable to non-transitory computer-readable media will be clearly visible in the detailed description provided below. It must be understood that the following detailed and specific embodiments are provided. When an exemplary embodiment of a method for predicting the popularity of a marketing target and a non-transitory computer-readable medium is proposed, it is only for the purpose of description and is not intended to limit the present invention. Range.
第1圖係顯示根據本發明一實施例所述之電子裝置之系統架構圖。系統架構100可實施於例如桌上型電腦、筆記型電腦、平板電腦或者智慧型手機等的電子裝置中,且至少包含一處理單元110。處理單元110可透過多種方式實施,例如以專用硬體電路或者通用硬體(例如,單一處理器、具平行處理能力之多處理器、圖形處理器或者其它具有運算能力之處理器),且於執行程式碼或者軟體時,提供之後所描述的功能。系統架構100更包括儲存單元120,用以儲存執行過程中所需要的資料以及各式各樣的電子檔案,例如各種演算法、參數、模型、文字檔和/或清單等。系統架構100更可包括網路介面130,用以下載上傳於網路上之文章或者文字檔等。顯示單元140可為顯示面板(例如,薄膜液晶顯示面板、有機發光二極體面板或者其它具顯示能力的面板),用以顯示輸入的字元、數字、 符號、拖曳鼠標的移動軌跡或者應用程式所提供的使用者介面,以提供給使用者觀看。此外,系統架構100更可包括一輸入單元150(例如滑鼠、觸控筆、鍵盤和/或觸控面板等),用以供使用者執行指令輸入等之操作。其中,顯示單元140與輸入單元150亦可結合為一觸控顯示面板,例如智慧型手機或者平板電腦之觸控顯示螢幕。值得注意的是,前述有關電子裝置之說明僅為一些示例,但本發明並不以此為限。 FIG. 1 is a system architecture diagram of an electronic device according to an embodiment of the present invention. The system architecture 100 may be implemented in an electronic device such as a desktop computer, a notebook computer, a tablet computer, or a smart phone, and includes at least one processing unit 110. The processing unit 110 may be implemented in a variety of ways, such as dedicated hardware circuits or general-purpose hardware (for example, a single processor, multiple processors with parallel processing capabilities, graphics processors, or other processors with computing capabilities), and When executing code or software, it provides the functions described later. The system architecture 100 further includes a storage unit 120 for storing data required during execution and various electronic files, such as various algorithms, parameters, models, text files, and / or lists. The system architecture 100 may further include a network interface 130 for downloading articles or text files uploaded on the network. The display unit 140 may be a display panel (for example, a thin-film liquid crystal display panel, an organic light-emitting diode panel, or other display-capable panel), and is used to display input characters, numbers, Symbols, dragging mouse movements, or the user interface provided by the app to provide users with a view. In addition, the system architecture 100 may further include an input unit 150 (such as a mouse, a stylus pen, a keyboard, and / or a touch panel, etc.) for the user to perform operations such as inputting instructions. The display unit 140 and the input unit 150 can also be combined into a touch display panel, such as a touch display screen of a smart phone or a tablet computer. It is worth noting that the foregoing description of the electronic device is only some examples, but the present invention is not limited thereto.
第2圖係顯示根據本發明一實施例所述之行銷標的熱門度預測方法之流程圖。於步驟S201,處理單元110透過網路介面130自社群網站下載對應一行銷類型之複數文章,並將其儲存至儲存單元120中。其中,行銷類型可包括美妝、家電、3C產品、生活用品、食品、運動用品等類型,但並不以此為限。於步驟S202,處理單元110透過斷詞(分詞)技術於文章中取得對應於該行銷類型之複數關鍵字。舉例來說,表1係顯示美妝領域所常見的關鍵字清單。 FIG. 2 is a flowchart illustrating a method for predicting popularity of a marketing target according to an embodiment of the present invention. In step S201, the processing unit 110 downloads a plurality of articles corresponding to a marketing type from a social network website through the network interface 130, and stores them into the storage unit 120. Among them, marketing types can include beauty, home appliances, 3C products, daily necessities, food, sports goods and other types, but it is not limited to this. In step S202, the processing unit 110 obtains plural keywords corresponding to the marketing type in the article through the word segmentation (word segmentation) technology. For example, Table 1 shows a list of keywords common in the beauty industry.
於步驟S203,取得關鍵字清單後,處理單元110根 據關鍵字清單中的所有關鍵字搜尋所有文章,並取得對應於每個關鍵字之時序資料。其中,取得時序資料之方式為先將所有文章以日、週、月、季或者年為單位進行分類,接著再統計每個關鍵字於複數文章中被提到的篇數。第3圖係顯示根據本發明一實施例所述之對應於不同關鍵字之時序資料之示意圖。於此實施例中,分類所有文章的時間單為係為一週,而縱軸則表示提到該關鍵字的文章篇數。舉例來說,如第3圖所示,關鍵字「蘭蔻」於第1週中被提到的文章篇數為13篇,而第2週中被提到的文章篇數則上升為16篇。關鍵字「雙色唇膏」於第1週中被提到的文章篇數為7篇,而第2週中被提到的文章篇數亦為7篇。關鍵字「唇釉」於第1週中被提到的文章中的篇數為6篇,而第2週中被提到的文章篇數仍為6篇,以此類推。其中,被提到的文章篇數越多,則代表該關鍵字的熱門度越高。 In step S203, after obtaining the keyword list, the processing unit 110 Search all articles based on all keywords in the keyword list and get time series data corresponding to each keyword. The way to obtain time series data is to first classify all articles by day, week, month, quarter, or year, and then count the number of articles mentioned in each article for each keyword. FIG. 3 is a schematic diagram showing timing data corresponding to different keywords according to an embodiment of the present invention. In this embodiment, the time list for classifying all articles is one week, and the vertical axis represents the number of articles mentioning the keyword. For example, as shown in Figure 3, the number of articles mentioned in the first week of the keyword "Lancome" was 13 and the number of articles mentioned in the second week rose to 16. The keyword "two-color lipstick" has 7 articles mentioned in the first week, and 7 articles mentioned in the 2nd week. The number of articles mentioned in the first week of the keyword "lip glaze" was six, while the number of articles mentioned in the second week was still six, and so on. Among them, the more articles mentioned, the higher the popularity of the keyword.
於步驟S204,於取得每個關鍵字的時序資料後,處理單元110更比較所有關鍵字中每兩個關鍵字之時序資料之波形,以決定每個關鍵字之間的相關性。計算時序資料之波形相關性之方式可包括動態時間校正(Dynamic Time Warping,DTW)、可將時間動態位移的皮爾森相關性分析(Pearson Correlation)和/或格蘭傑因果關係(Granger Causality)等,但並不以此為限。其中,前述的計算方式將允許處理單元110壓縮或者擴張對應於一較小的時間區段之波形,或者對波形進行偏移(shift),以與另一關鍵字之波形進行比對,來決定兩個關鍵字之間的相關性。舉例來說,若兩個時間點所構成的波形與三個時間點所構成的波形相似,則經過處理單元110對波 形進行壓縮或者擴張後,及可判斷兩個波形類似。其中,計算波型相似度之公式如下所示:D(i,j)=|t(i)-r(j)|+min{(D(i-1,j),D(i-1,j-1),D(i,j-1))}其中,|t(i)-r(j)|為兩序列數值相減所得到的路徑成本。然而,當兩個關鍵字(i,j)之距離成本D小於一預設的既定值,則處理單元110即判斷該兩個關鍵字具有關聯性。 In step S204, after obtaining the time series data of each keyword, the processing unit 110 further compares the waveforms of the time series data of each of the two keywords to determine the correlation between each keyword. Methods for calculating the correlation of waveforms of time series data may include Dynamic Time Warping (DTW), Pearson Correlation analysis that can dynamically shift time, and / or Granger Causality, etc. , But not limited to this. The foregoing calculation method will allow the processing unit 110 to compress or expand a waveform corresponding to a smaller time period, or shift the waveform to compare it with a waveform of another keyword to determine Relevance between two keywords. For example, if the waveform formed at two time points is similar to the waveform formed at three time points, after the waveform is compressed or expanded by the processing unit 110, it can be determined that the two waveforms are similar. Among them, the formula for calculating the wave shape similarity is as follows: D ( i , j ) = | t ( i ) -r ( j ) | + min {(D ( i -1, j ), D ( i -1, j -1), D ( i , j -1))} where | t ( i ) -r ( j ) | is the path cost obtained by subtracting the two sequence values. However, when the distance cost D of the two keywords (i, j) is less than a preset predetermined value, the processing unit 110 judges that the two keywords are related.
此外,由於某些特定關鍵字於文章中被提到的篇數過多或者過少,使得儘管其波形與其它關鍵字之波形類似,但因為其篇數的數量與其它關鍵字差異過大而無法正確地找出兩個關鍵字之間之相關性。因此,根據本發明另一實施例,於取得每個關鍵字之時序資料後,處理單元110更對每個時序資料進行正規化,以避免差異過大的情況產生。舉例來說,於正規化之前,雙色唇膏的時序資料為(138,141,152,165,173,313),而金高恩的時序資料為(75,51,38,47,15,28)。而根據上述之時序資料可得知,由於雙色唇膏與金高恩的時序資料之間之差異非常大,因此儘管兩者之波形相似,處理單元110可能無法判斷出兩者具有相關性。而經過正規化之處理後,雙色唇膏的時序資料變成(0.0,1.7,8.0,15.4,20.0,100.0),而金高恩的時序資料則變成(100.0,60.0,38.3,53.3,0.0,21.7)。相較於進行正規化處理前之數據,雙色唇膏以及金高恩之時序資料之間之差異已經變得較小,將有利於處理單元110執行相關性之判斷。其中,執行正規化之公式如下所示:
當處理單元110判斷某兩個關鍵字具有關聯性時,進入步驟S205,處理單元110接著判斷兩個關鍵字之領先關係。其中,領先時序的判斷可使用以下的公式來計算:
第4圖係顯示根據本發明一實施例所述之關鍵字「香氛」以及關鍵字「SK-II」之間之領先關係之示意圖。如第4圖所示,根據計算之結果,關鍵字「香氛」之時序係領先於關鍵字「SK-II」之時序。 FIG. 4 is a schematic diagram showing a leading relationship between a keyword “fragrance” and a keyword “SK-II” according to an embodiment of the present invention. As shown in Figure 4, according to the calculation results, the timing of the keyword "fragrance" is ahead of the timing of the keyword "SK-II".
於取得每兩個關鍵字之間之領先關係後,進入步驟S206,處理單元110根據所有關鍵字之間之領先關係建立一時序關聯矩陣。舉例來說,第5圖係顯示根據本發明一實施例所述之時序關聯矩陣之示意圖。於此實施例中,於前述之步驟 S205之計算結果可得知,關鍵字「蘭蔻」、「歐蕾」、「沐浴」、「碧兒泉」、「凝露」、「沐浴乳」、「Facial」等之時序係領先關鍵字「雙色唇膏」之時序,關鍵字「蘭蔻」、「歐蕾」、「凝露」、「嫩白」、「全效」、「Facial」、「噴霧」、「碧兒泉」、「雙色唇膏」等之時序係領先關鍵字「唇釉」之時序,而根據上述之領先關係即可得到如第5圖所示之內容。 After obtaining the leading relationship between each two keywords, proceed to step S206, and the processing unit 110 establishes a time-series association matrix according to the leading relationship between all keywords. For example, FIG. 5 is a schematic diagram showing a timing correlation matrix according to an embodiment of the present invention. In this embodiment, in the foregoing steps According to the calculation result of S205, the keywords "Lancome", "Ole", "bathing", "Bierquan", "gel", "bathing milk", "Facial", etc. are the leading keywords " The timing of "two-color lipstick", keywords "lancome", "Ole", "dew", "whitening", "full effect", "Facial", "spray", "Bierquan", "two-color lipstick" The timing of waiting is the timing of the leading keyword "lip glaze", and the content shown in Figure 5 can be obtained according to the above leading relationship.
於步驟S207,處理單元110將每個關鍵字的時序資料以及每兩個關鍵字之間之領先關係(即時序關聯矩陣)作為訓練資料,並透過一遞歸神經網絡(Recurrent Neural Network,RNN)(例如長短期記憶(Long Short-Term Memory,LSTM),但並不以此為限)等建立一預測模型,但並不以此為限。 In step S207, the processing unit 110 uses the time series data of each keyword and the leading relationship between each two keywords (i.e., the time series correlation matrix) as training data, and passes a Recurrent Neural Network (RNN) ( For example, Long Short-Term Memory (LSTM), but not limited to it, is used to build a prediction model, but it is not limited to this.
於步驟S208,使用者於關鍵字清單中選擇一關鍵字作為預測詞彙,並將其輸入至預測模型中,以產生對應於該預測詞彙之一推薦清單。其中,處理單元110首先根據時序關聯矩陣找出與預測詞彙相關且被預測詞彙領先的多個關鍵字(即找出預測詞彙可能會帶動的關鍵字),接著將預測詞彙以及多個關鍵字輸入至預測模型中以取得預測的熱門度,並給予該些關鍵字對應之權重以供使用者作為選擇之依據。舉例來說,於此實施例中,使用者係選擇「雙色唇膏」作為預測詞彙。根據第5圖所示之時序關聯矩陣,與「雙色唇膏」具有相關性且被「雙色唇膏」領先的關鍵字包括「唇釉」以及「唇彩」,表示「唇釉」以及「唇彩」可能於未來會成為與「雙色唇膏」相關的熱門關鍵字。此外,根據第5圖所示之內容可得知,「控油」、「滋潤」等關鍵字與「唇釉」相關(即兩者之時間序列波 形相似),以及「噴霧」、「膠原」等關鍵字則與「唇彩」相關,且上述之關鍵字皆分別領先「唇釉」以及「唇彩」。因此,處理單元110將「雙色唇膏」與「控油」、「滋潤」等關鍵字作為對應於「唇彩」之輸入以及將「雙色唇膏」與「噴霧」、「膠原」等關鍵字作為對應於「唇釉」之輸入,以自預測模型中取得「唇釉」以及「唇彩」於未來時間點的聲量(Voice)。接著,處理單元110即根據「唇釉」以及「唇彩」所分別對應之聲量給予不同之推薦權重,以產生推薦清單。舉例來說,於此實施例中,根據預測模型之計算結果推得「唇釉」所對應之權重為0.3,而「唇彩」所對應之權重則為0.8,表示「唇彩」於未來時間點較「唇釉」熱門。 In step S208, the user selects a keyword from the keyword list as a predicted vocabulary and inputs it into the prediction model to generate a recommended list corresponding to the predicted vocabulary. Among them, the processing unit 110 first finds a plurality of keywords related to the predicted vocabulary and being ahead of the predicted vocabulary according to the time-series association matrix (that is, finds keywords that the predicted vocabulary may drive), and then inputs the predicted vocabulary and multiple keywords To the prediction model to obtain the predicted popularity, and give the keywords corresponding weights for users to use as a basis for selection. For example, in this embodiment, the user selects "two-color lipstick" as the predicted vocabulary. According to the timing correlation matrix shown in Figure 5, the keywords that are related to "two-color lipstick" and are led by "two-color lipstick" include "lip glaze" and "lip gloss", indicating that "lip glaze" and "lip gloss" may be used in It will become a popular keyword related to "two-color lipstick" in the future. In addition, according to the content shown in Figure 5, it can be known that the keywords "oil control" and "moisturizing" are related to "lip glaze" (that is, the time series wave of the two) And similar keywords), and keywords such as "spray" and "collagen" are related to "lip gloss", and the above keywords are respectively ahead of "lip glaze" and "lip gloss". Therefore, the processing unit 110 uses keywords such as “two-color lipstick” and “oil control” and “moisturizing” as inputs corresponding to “lip gloss” and uses keywords such as “two-color lipstick” and “spray” and “collagen” as corresponding to “ "Lip glaze" is used to obtain the "Voice" of "lip glaze" and "lip gloss" at a future point in time from the prediction model. Then, the processing unit 110 assigns different recommendation weights according to the respective sound volumes corresponding to "lip glaze" and "lip gloss" to generate a recommendation list. For example, in this embodiment, according to the calculation result of the prediction model, the weight corresponding to "lip glaze" is 0.3, and the weight corresponding to "lip gloss" is 0.8, which means that "lip gloss" will "Lip glaze" is hot.
於步驟S209,處理單元110根據推薦清單中每個預測關鍵字所對應之權重決定對應於預測詞彙之一行銷標的。舉例來說,根據上述之實施例,由於「唇彩」所對應之權重大於「唇釉」所對應之權重,因此處理單元110選擇具有較大權重的「唇彩」作為行銷標的。 In step S209, the processing unit 110 determines a marketing target corresponding to one of the predicted words according to the weight corresponding to each predicted keyword in the recommendation list. For example, according to the above-mentioned embodiment, since the weight corresponding to "lip gloss" is greater than the weight corresponding to "lip glaze", the processing unit 110 selects "lip gloss" with a larger weight as the marketing target.
根據本發明另一實施例,隨著不同產品的開發,當有新的明星代言新的產品時,新的文章中可能會出現新的關鍵字。而為了找出新的關鍵字,處理單元110可透過計算未知詞彙(Out-of-vocabulary word)與現有的關鍵字於文章中共同出現的頻率,並同時計算未知詞彙出現於不同行銷類型之數量,以判斷該未知詞彙是否為新的關鍵字。其中,新的關鍵字的判斷可透過以下的公式來進行:
此外,在上述示例性裝置中,儘管上述方法已在使用一系列步驟或方塊之流程圖的基礎上描述,但本發明不侷限於這些步驟的順序,且一些步驟可不同於其餘步驟的順序執行或者其餘步驟可同時進行。以及,本領域的技術人士將理解在流程圖中所示的步驟並非唯一的,其可包括流程圖的其他步驟,或者一或多個步驟可被刪除而不會影響本發明之範圍。 In addition, in the above exemplary apparatus, although the above method has been described on the basis of a flowchart using a series of steps or blocks, the present invention is not limited to the order of these steps, and some steps may be performed differently from the order of the remaining steps Or the remaining steps can be performed simultaneously. And, those skilled in the art will understand that the steps shown in the flowchart are not unique, they may include other steps of the flowchart, or one or more steps may be deleted without affecting the scope of the present invention.
綜上所述,根據本發明一實施例所提出之行銷標的熱門度預測方法以及非暫態電腦可讀取媒體,使用者可藉由前述的預測模型來預測與預測詞彙相關的複數關鍵字於未來可能的熱門情況,以提早進行廣告投遞以及行銷策略的相關佈局等,使得於相同的行銷預算下,藉由將預算投入至價格較低但具有發展性的行銷標的以得到更高的行銷效益。 In summary, according to a method for predicting the popularity of marketing targets and a non-transitory computer-readable medium according to an embodiment of the present invention, a user can predict the plural keywords related to the predicted vocabulary through the aforementioned prediction model. Possible future hot situations, such as early advertising delivery and related layout of marketing strategies, so that under the same marketing budget, the budget can be invested into lower-priced but developmental marketing targets to obtain higher marketing benefits. .
以上敘述許多實施例的特徵,使所屬技術領域中具有通常知識者能夠清楚理解本說明書的形態。所屬技術領域中具有通常知識者能夠理解其可利用本發明揭示內容為基礎以設計或更動其他製程及結構而完成相同於上述實施例的目的及/或達到相同於上述實施例的優點。所屬技術領域中具有通常知識者亦能夠理解不脫離本發明之精神和範圍的等效構造可在不脫離本發明之精神和範圍內作任意之更動、替代與潤飾。 The features of many embodiments described above enable those skilled in the art to clearly understand the form of this specification. Those skilled in the art can understand that they can use the disclosure of the present invention as a basis to design or modify other processes and structures to accomplish the same purpose and / or achieve the same advantages as the above embodiments. Those with ordinary knowledge in the technical field can also understand that equivalent structures without departing from the spirit and scope of the present invention can be arbitrarily changed, substituted, and retouched without departing from the spirit and scope of the present invention.
Claims (20)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW106145044A TWI660317B (en) | 2017-12-21 | 2017-12-21 | Methods for predicting marketing target popularity and non-transitory computer-readable medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW106145044A TWI660317B (en) | 2017-12-21 | 2017-12-21 | Methods for predicting marketing target popularity and non-transitory computer-readable medium |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI660317B true TWI660317B (en) | 2019-05-21 |
TW201928831A TW201928831A (en) | 2019-07-16 |
Family
ID=67347998
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW106145044A TWI660317B (en) | 2017-12-21 | 2017-12-21 | Methods for predicting marketing target popularity and non-transitory computer-readable medium |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI660317B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103258000A (en) * | 2013-03-29 | 2013-08-21 | 北界创想(北京)软件有限公司 | Method and device for clustering high-frequency keywords in webpages |
TW201539212A (en) * | 2014-04-11 | 2015-10-16 | Univ Chia Nan Pharm & Sciency | Construction method, database system, computer program, and computer readable recoding medium of relevant literature inquiry |
US20160224662A1 (en) * | 2013-07-17 | 2016-08-04 | President And Fellows Of Harvard College | Systems and methods for keyword determination and document classification from unstructured text |
CN107480822A (en) * | 2017-08-14 | 2017-12-15 | 国云科技股份有限公司 | A kind of marketing enterprises development trend Forecasting Methodology based on TrieTree |
-
2017
- 2017-12-21 TW TW106145044A patent/TWI660317B/en active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103258000A (en) * | 2013-03-29 | 2013-08-21 | 北界创想(北京)软件有限公司 | Method and device for clustering high-frequency keywords in webpages |
US20160224662A1 (en) * | 2013-07-17 | 2016-08-04 | President And Fellows Of Harvard College | Systems and methods for keyword determination and document classification from unstructured text |
TW201539212A (en) * | 2014-04-11 | 2015-10-16 | Univ Chia Nan Pharm & Sciency | Construction method, database system, computer program, and computer readable recoding medium of relevant literature inquiry |
CN107480822A (en) * | 2017-08-14 | 2017-12-15 | 国云科技股份有限公司 | A kind of marketing enterprises development trend Forecasting Methodology based on TrieTree |
Also Published As
Publication number | Publication date |
---|---|
TW201928831A (en) | 2019-07-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11416765B2 (en) | Methods and systems for evaluating training objects by a machine learning algorithm | |
US20230229692A1 (en) | Image-based popularity prediction | |
CN107315759B (en) | Method, device and processing system for classifying keywords and classification model generation method | |
CN106650780B (en) | Data processing method and device, classifier training method and system | |
US8880441B1 (en) | Click stream analysis for fraud detection | |
US20200110842A1 (en) | Techniques to process search queries and perform contextual searches | |
WO2018166288A1 (en) | Information presentation method and device | |
WO2017080170A1 (en) | Group user profiling method and system | |
JP2016529611A (en) | Method and system for retrieving images | |
JP2016531350A (en) | Method and system for recommending online products | |
Hsu et al. | Flexible dynamic time warping for time series classification | |
Anwar et al. | Hand gesture recognition: a survey | |
Jiang et al. | Online robust action recognition based on a hierarchical model | |
WO2017214970A1 (en) | Building convolutional neural network | |
JP7160843B2 (en) | Prediction of molecular properties of molecular variants using residue-specific structural features | |
Li et al. | Weighted doubly regularized support vector machine and its application to microarray classification with noise | |
WO2020155814A1 (en) | Damage determination method and apparatus for maintenance object, and an electronic device | |
Bai et al. | Cross-domain representation learning by domain-migration generative adversarial network for sketch based image retrieval | |
TWI660317B (en) | Methods for predicting marketing target popularity and non-transitory computer-readable medium | |
TWI704804B (en) | Electronic apparatus and automatic advertisement closing method thereof | |
US9576203B2 (en) | Devices, systems, and methods for knowledge-based inference for material recognition | |
JP6592574B1 (en) | Article analysis apparatus and article analysis method | |
Gudmundsson et al. | Exploring Sub-skeleton Trajectories for Interpretable Recognition of Sign Language | |
Tyagi et al. | Hand Anatomy and Neural Network-Based Recognition for Sign Language | |
US12033411B2 (en) | Stroke based control of handwriting input |